I have another stack of sixgill changes built up. I'm just going to post the changes that I think are useful to review. I have other stuff that doesn't seem worth chewing up review time that I plan to build with but not request review for. These include: - making it compatible with gcc 4.9 - getting rid of a bunch of warnings - statically linking with libgmp (required for several gcc versions) - building without cvc3 (to avoid the dependency) - adding the ability for xmanager to log and replay all input from plugins - some fixes to the memory tracking - improved a couple of error messages I suppose I should put up a repo on github or somewhere for these.

Handle sizeof and alignof. sizeof seems to be the largest source of unhandled constructs currently. This is totally cargo-culted.

This implements a bare-bones garbage collector, relying on the existing marking infrastructure. I don't really have any clue about the data structures and in particular where all the roots are hiding, so I just added stuff until it stopped crashing. This seems to free a large number of objects while producing bitwise identical .xdb files (at least for a compiling a single unified file). I don't know whether it really reduces memory usage or not. (Hmm... come to think of it, I bet I should be doing UnPersist before deleting...)

Well, it doesn't crash with the UnPersist added. Must be correct. :-)

Comment on attachment 8569552 [details] [diff] [review] Handle sizeof and alignof Review of attachment 8569552 [details] [diff] [review]: ----------------------------------------------------------------- This patch just seems to be translating sizeof(e) into just e, i.e. it passes through the operand. I think you would need to be using TREE_TYPE, TYPE_SIZE, and XIL_ExpInt to get the type of the expression and compute a constant for its size.

Ok, let me try this again. More cargo-culting.

When I implemented name mangling, I just freely leaked strings all over. As long as I'm fixing memory problems, I should really clean them up. This really wants RAII, but it's a quick fix.

Comment on attachment 8570851 [details] [diff] [review] Stop leaking strings Review of attachment 8570851 [details] [diff] [review]: ----------------------------------------------------------------- ::: gcc/xgill.c @@ +70,5 @@ > + > + XIL_ReleaseCString(dst); > + dst->str = src->str; > + dst->owned = src->owned; > + src->owned = 0; false

I have deployed builds with all these changes.